rknn_model_zoo，运行同样的程序，为什么会得到不同的结果

神奇的潘子

平台是 RK3576
模型是 自己训练的yolov8_seg模型
配置参数
#define OBJ_CLASS_NUM 3
#define NMS_THRESH 0.6
#define BOX_THRESH 0.6



担心图看不清，上输出
cat@lubancat:~/rknn/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov8_seg_demo$ ./rknn_yolov8_seg_demo ./model/yolov8_seg.rknn ~/20250711133825.jpg
load lable ./model/coco_80_labels_list.txt model input num: 1, output num: 13, height=640, width=640, channel=3
fill dst image (x y w h)=(0 0 640 640) with color=0x72727272
rga_api version 1.10.1_[0]
-- matmul_by_cpu_uint8 use: 10.290000 ms
-- resize_by_opencv_uint8 use: 1.154000 ms
-- crop_mask_uint8 use: 1.359000 ms
-- seg_reverse use: 1.601000 ms
commodity @ (272 170 315 243) 0.655
write_image path: out.png width=640 height=480 channel=3 data=0x55b4e6cb40
cat@lubancat:~/rknn/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov8_seg_demo$ ./rknn_yolov8_seg_demo ./model/yolov8_seg.rknn ~/20250711133825.jpg
load lable ./model/coco_80_labels_list.txt model input num: 1, output num: 13, height=640, width=640, channel=3
fill dst image (x y w h)=(0 0 640 640) with color=0x72727272
rga_api version 1.10.1_[0]
-- matmul_by_cpu_uint8 use: 18.888000 ms
-- resize_by_opencv_uint8 use: 1.983000 ms
-- crop_mask_uint8 use: 2.680000 ms
-- seg_reverse use: 1.619000 ms
commodity1 @ (272 170 315 243) 0.781
commodity @ (272 169 315 242) 0.633
write_image path: out.png width=640 height=480 channel=3 data=0x55be91cb40
cat@lubancat:~/rknn/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov8_seg_demo$ ./rknn_yolov8_seg_demo ./model/yolov8_seg.rknn ~/20250711133825.jpg
load lable ./model/coco_80_labels_list.txt model input num: 1, output num: 13, height=640, width=640, channel=3
fill dst image (x y w h)=(0 0 640 640) with color=0x72727272
rga_api version 1.10.1_[0]
-- matmul_by_cpu_uint8 use: 6.526000 ms
-- resize_by_opencv_uint8 use: 0.878000 ms
-- crop_mask_uint8 use: 1.226000 ms
-- seg_reverse use: 1.447000 ms
commodity @ (272 170 315 243) 0.655
write_image path: out.png width=640 height=480 channel=3 data=0x55881acb40
cat@lubancat:~/rknn/rknn_model_zoo/install/rk3576_linux_aarch64/rknn_yolov8_seg_demo$ ./rknn_yolov8_seg_demo ./model/yolov8_seg.rknn ~/20250711133825.jpg
load lable ./model/coco_80_labels_list.txt model input num: 1, output num: 13, height=640, width=640, channel=3
fill dst image (x y w h)=(0 0 640 640) with color=0x72727272
rga_api version 1.10.1_[0]
-- matmul_by_cpu_uint8 use: 14.192000 ms
-- resize_by_opencv_uint8 use: 1.968000 ms
-- crop_mask_uint8 use: 2.679000 ms
-- seg_reverse use: 1.627000 ms
commodity1 @ (272 170 315 243) 0.778
commodity @ (272 169 315 242) 0.633
write_image path: out.png width=640 height=480 channel=3 data=0x55a021eb40

神奇的潘子

自定义的模型类的数量与程序的类的模型数量不一致导致的。